MedlinePlus Connect: Planning for Clinical Coding System Changes

LISTER HILL CENTER RECOMMENDATIONS

The next step of the project is to review the mappings that result from the LHC-UMLS algorithm. It is recommended that the team take time to review the latest algorithm output from LHC and determine patterns that can be used to refine the mapping. In addition, several questions can be answered during the mapping review, as shown below:

1. 
   How does a random sampling of the LHC-UMLS mappings compare with the GEMs random sampling in terms of ICD-10-CM overall coverage and chapters with issues (Forward mapping GEMs and their distribution according to chapter and number of issues per chapter is available in Table 9)? Do particular chapters have fewer mapping issues than others? Do particular chapters have more mapping issues than others?
   
2. 
   Which mapping methods (I, A, G/x, O) provide correct results that will help MedlinePlus Connect support the next version of the International Classification of Disease, ICD-10-CM? Can any of the mapping methods be improved to help MedlinePlus Connect support ICD-10-CM?
   
3. 
   How can the LHC-UMLS algorithm best help MedlinePlus Connect in the long term or will the algorithm need to be revised with each major ICD revision?
   
4. 
   Are there patterns to how the LHC algorithm maps ICD-10-CM to the Health Topics?
   
5. 
   Will the LHC-UMLS mappings provide an easier method to support ICD-10-CM than the GEMs in the short term? Long term? Should a combination method of GEMs and LHC mappings be used to support ICD-10-CM?
   

In addition, the Associate Fellow recommends that the MedlinePlus Connect team takes several additional steps. The team can create a blacklist of MedlinePlus Health Topics and ICD-10-CM codes that can be applied as a filter to the algorithm and continue working with Anna Ripple, Olivier Bodenreider, and Lee Peters to determine other methods to improve the algorithm or create filters. Lee Peters developed a graphical user interface that will allow the MedlinePlus Connect team to better analyze how the algorithm creates the resulting mappings. The team can determine if the graphical user interface developed by Lee Peters will help analyze and improve the mappings. For instance, can certain sources be removed that provide particular incorrect mappings?

Once these steps have been taken, the team must ultimately decide whether or not the LHC-UMLS mappings and/or the GEMs should be used to support ICD-10-CM. [Note: the next release of ICD-9-CM and ICD-10-CM occurs October 1, 2011. New GEMs will be released at this time. Once these GEMs are released, the MedlinePlus Connect team can begin the mapping process if the GEMs are chosen as part or all of the approach].

GENERAL RECOMMENDATIONS

Therefore, in addition to the GEMs and LHC specific recommendations, several broad recommendations for how the project should move forward are discussed below:

1. 
   In the immediate future, the team must finish the LHC-UMLS mapping evaluation and compare a sample of these to the GEMs sample. Rex Robison is currently on detail from the NIH Library with MedlinePlus Connect. He will finish this last analysis and comparison.
   
2. 
   The team can then decide on a mapping method. There are several options:
   
   • 
     Option 1: Use either the GEMs or LHC-UMLS mappings by themselves to provide the baseline mappings and manually review each.
     
   • 
     Option 2: Choose one method as the primary mapping method for the transition from ICD9 to 10CM and to use the other method for review and quality assurance. With the significant increase in codes, additional review is recommended.
     
   • 
     Option 3: Use a combination of both methods. For instance, the team could use the GEMs in combination with mappings that results from the closest relationships within the UMLS.
     
3. 
   While there are still over two years until the transition to ICD-10-CM, the last recommendation is for the team to decide which mapping option to use by the end of the year and certainly after the October release of ICD-9-CM and ICD-10-CM. Plan a significant amount of time for the mapping process and to allow adequate time to complete and review these mappings so that MedlinePlus Connect is prepared to support clinical system testing prior to October 2013.
   

The Associate Fellow also recommends that the team determine exactly how the MedlinePlus Health Topics are represented within the UMLS. While this is potentially a longer term project, determining this could potentially help improve the LHC-UMLS algorithm results while also providing a more “true” version of the Health Topics within the UMLS.

In addition, the NLM is working to provide a mapping between the SNOMED-CT Problem List Subset and ICD-10-CM. Once completed, these mappings could potentially provide the team with another method to support ICD-10-CM and it is recommended that the team consider how these mappings can benefit MedlinePlus Connect once they are released.

DISCUSSION

Using the GEMs mapping suggestions or LHC’s mapping will both require manual review. Fortunately, the change will not go into effect until October 2013 and there are a number of ways that MedlinePlus Connect can begin automating portions of mappings from each file and expanding the number of ICD-10-CM codes and Health Topic matches manually.

Using the GEMs will require significant manual review but the MedlinePlus Connect team can work with a programmer to automate large portions of the work. The benefit of the GEMs is they provide mapping suggestions to concepts that were previously mapped in ICD-9-CM and can build upon previous work of the MedlinePlus Connect team. Yet, this means that unmapped ICD-9-CM codes and the corresponding ICD-10-CM codes will not have a Health Topic associated. To address this issue, the team can work with the descendents or siblings of currently mapped concepts to expand the list. A potential negative about using the GEMs is they do not help MedlinePlus Connect expand the number of mappings available between ICD-10-CM and the Health Topics since they only work with codes that can be used for reimbursement, and not more general codes that provide context within ICD-10-CM but cannot be used for billing. Several institutions have noted that they use the ICD-9-CM codes that are not used for reimbursement internally and that MedlinePlus Connect mappings to these codes are useful. The MedlinePlus Connect team can decide whether or not to map to these codes. If the mappings are desired, they can potentially be automated by using suggestions from code pairs lower down the hierarchy. In addition, if the MedlinePlus Connect team uses the GEMs and reviews the master file of forward and backward mappings, all ICD-9-CM and ICD-10-CM concepts will be included. Instances where an ICD-10-CM code has multiple Health Topics will need to be reviewed as previously mentioned in order to maintain a consistent file. Using the master file will help the MedlinePlus Connect team view all possible health topic suggestions for each ICD-10-CM code from both the forward and backward mapping file. While not a perfect solution, the GEMs do appear to provide a good base for the MedlinePlus Connect team to support ICD-10-CM.

The LHC’s algorithm’s mappings provide a second option for MedlinePlus Connect. While the algorithm does not perpetuate issues from the initial MedlinePlus mapping work with ICD-9-CM, it introduces new issues that will also require manual review by the MedlinePlus Connect team. This algorithm provides mappings to ICD-10-CM codes throughout the hierarchy and not only codes used for billing and reimbursement. Yet, the mapping does not include all billable codes and the team would need to work with the unmapped siblings and children of mapped codes to increase the number of Health Topic mappings. Other benefits or issues with the LHC algorithm remain to be seen.